Maximization of strength–ductility balance of dual-phase steels using generative adversarial networks and Bayesian optimization

Abstract

Dual-phase (DP) steels with a soft ferrite matrix and hard martensite islands exhibit an excellent combination of strength and ductility. However, further improvement of the strength–ductility balance is desirable for automotive applications, and the optimization of the DP microstructure is crucial for enhancing mechanical properties. This study aimed to maximize the strength–ductility balance of DP steel using an integrated computational framework comprised by generative adversarial networks (GAN), finite element method (FEM), and Bayesian optimization. GAN was trained on the microstructures of real DP steels to generate synthetic microstructure images randomly mapped from latent variable vectors, and the tensile properties of the generated microstructures were evaluated using FEM. Bayesian optimization was then employed to identify latent variables yielding microstructures with the highest tensile strength (TS), uniform elongation (uEL), or strength–ductility balance (TS × uEL). The optimized microstructures of DP steels were then quantitatively analyzed to elucidate the relationship between the microstructural morphology and tensile properties. The optimal synthetic DP microstructure exhibited a TS × uEL value higher by 4,027MPa% than the highest value shown by real DP steels. Thus, the proposed integrated optimization framework enables efficient computational design of various material microstructures for maximizing desired mechanical properties.